[ECOS] Network latency
Mon Oct 26 16:29:00 GMT 2009
I’m currently experiencing some strange behaviour with network latency.
We’ve got 2 boxes, both running eCos connected via Ethernet, and every
so often we get a lag in message delivery from a user pressing a H/W
switch on one box, which sends a msg via TCP to the other. Roughly 2
or 3 seconds.
I’ve gone back to the nc_master / slave tests and monitored the
throughput when connecting a PC to one of the boxes running eCos.
Initially everything looked fine until I insert a switch between the
PC and eCos box.
With any switch between the PC / eCos box, there is poor performance
when doing an RX test (PC -> eCos box), which shows lots of out of
order delivery / delayed ACK / fast rexmit, and the throughout graph
looks like a saw tooth.
But hey, the PC’s running faster than the eCos box, so maybe this is
to be expected.
So I’ve gone back to the eCos -> eCos scenario and extern’d some of
the netstack stats and I find there is a direct correlation between
the following stats increasing and the delivery delay:
Everything will be nice and stable, with consistent message delivery
times, then suddenly one of the three above fires and I get 2 -> 4
second delay of message delivery.
I’m wondering if the switches I’m placing between the units are
causing some sort of flow control issue ?
Does anyone know if in the PHY drivers when performing the auto
negotiation configures any flow control with a switch, and if this is
likely to be the cause?
I’m also playing with TCP_NODELAY which seems to influence the
behaviour, but doesn’t completely fix the issue.
Any help would be appreciated.
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
More information about the Ecos-discuss